This paper introduces SoccerNet-GSR, a novel dataset and evaluation metric for Game State Reconstruction (GSR) in sports. GSR aims to track and identify all athletes on a sports pitch based on input videos captured by a single camera. The dataset includes 200 30-second video sequences annotated with 9.37 million line points for pitch localization and camera calibration, as well as over 2.36 million athlete positions with their respective role, team, and jersey number. A new evaluation metric, GS-HOTA, is introduced to assess GSR methods. Additionally, an end-to-end baseline for GSR is proposed and released. The paper also discusses the challenges of GSR, including the need for accurate athlete localization and identification, and the importance of combining these tasks in a single pipeline. The GSR-Baseline is evaluated on the SoccerNet-GSR dataset, showing that it achieves a GS-HOTA score of 22.26% on the test set. The results highlight the complexity of the GSR task and the importance of accurate calibration and athlete identification. The paper concludes that the introduction of this benchmark opens up new research opportunities in the field of sports analytics.This paper introduces SoccerNet-GSR, a novel dataset and evaluation metric for Game State Reconstruction (GSR) in sports. GSR aims to track and identify all athletes on a sports pitch based on input videos captured by a single camera. The dataset includes 200 30-second video sequences annotated with 9.37 million line points for pitch localization and camera calibration, as well as over 2.36 million athlete positions with their respective role, team, and jersey number. A new evaluation metric, GS-HOTA, is introduced to assess GSR methods. Additionally, an end-to-end baseline for GSR is proposed and released. The paper also discusses the challenges of GSR, including the need for accurate athlete localization and identification, and the importance of combining these tasks in a single pipeline. The GSR-Baseline is evaluated on the SoccerNet-GSR dataset, showing that it achieves a GS-HOTA score of 22.26% on the test set. The results highlight the complexity of the GSR task and the importance of accurate calibration and athlete identification. The paper concludes that the introduction of this benchmark opens up new research opportunities in the field of sports analytics.